Gary Marcus

Rank 13 of 47

Score 84

Update: o3 appears to me (baed on informal testing) to be less likely to succumb to simple “potemkins” but not immune.

Here (left) it affirms what a haiku is, but tells me (wrongly) that “asphalt shimmers” fits the requirement that the last line of haiku should be five

6/27/2025, 7:01:38 PM

In reply to:

Gary Marcus

@GaryMarcus

229d

BREAKING: Explosive new paper from MIT/Harvard/UChicago.

Things just got worse — a lot worse — for LLM’s and the myth that they can understand and reason.

The paper documents a pattern they called Potemkins, a kind of reasoning inconsistency (see figure below). They show that

The statement and accompanying images discuss the limitations of large language models (LLMs) in understanding and reasoning, specifically highlighting a phenomenon called 'Potemkin understanding.' The statement critiques the models' ability to adhere to logical consistency, using examples of haiku syllable counting and rhyming schemes to illustrate these shortcomings.

Principle 1:
I will strive to do no harm with my words and actions.
The statement aims to highlight potential issues with LLMs, which could contribute to improving their development, aligning with the principle of doing no harm. [+1]
Principle 3:
I will use my words and actions to promote understanding, empathy, and compassion.
By discussing the limitations of LLMs, the statement promotes understanding of their current capabilities and limitations. [+1]
Principle 5:
I will acknowledge and correct my mistakes.
The statement acknowledges and corrects a mistake in the syllable count, demonstrating a commitment to accuracy. [+1]